Frame Burst Adjusting for Transmitting Video Conference in Gigabit Ethernet

Frame Burst Adjusting for Transmitting Video Conference in Gigabit Ethernet Han-Chieh Chao and Yao-Chung Chang Institute of Electrical Engineering National Dong Hwa University Hualien, Taiwan E-mail: hcc@cc.ndhu.edu.tw Tel: +886-3-8662500 ext.1501 Fax: +886-3-8662509 Abstract Due to the development of the network techniques is so fast, the bandwidth of Ethernet starting form 10Mbps, 100Mbps comes to 1000Mbps. The basic issue is to keep the compatibility between three kind of different speeds Ethernet networks. Carrier Extension is adopted to solve the different size of frames in Gigabit Ethernet, fast Ethernet, and standard Ethernet. Transmitting several frames within one burst will increase the overall throughput of Gigabit Ethernet. In this paper we will transmit real time application such as Video Conference over 1000Mbps network. Using Frame Burst to combine several packets within one period of transmission time will result a better throughput, also the delay of frames will change due to the burst length. Finally, we simulate the outcome of transmitting Video Conference application, and discuss the main factors effect the performance of Video Conference under gigabit Ethernet. At the low traffic load such as 15%, burst length does not effect network traffic. Increasing the traffic load, the effects of changing burst length are clear to observe. When transmitting Video Conference, it has the minimum access delay at the burst length 131072BT traffic load is 60%. FB2.0 while the Key Words: Gigabit, Ethernet, Frame Burst, Bit Time. 1. Introduction Today, among the high speed LAN technologies like Fast Ethernet or 100Base-T can provide greater bandwidth and improve the client/server response time. Fast Ethernet technology provides a smooth, non-disruptive evolution to 100-Mbps performance. Using 100Base-T connections between servers and desktop is becoming a tendency for connecting backbone and servers. Basically, those technologies should provide a scalable upgrade path, cost effective and minimum retraining. The recent advance in video coding emerges the coming of Gigabit Ethernet. The increasing of data rate and characteristic make it better performance to transmit real time applications. Using more bandwidth of Gigabit Ethernet to solve the original LAN poor bandwidth is a simple solution. But it still needs further improvement. 2. Gigabit Ethernet In July 1996, the IEEE 802.3 working group created the 802.3z Gigabit Ethernet task force. The major object of the 802.3z Gigabit Ethernet task force are to develop a Gigabit Ethernet standard that follow below Allows both half-duplex and full-duplex operation at 1000Mbps speed. Uses the same frame format with IEEE 802.3 Ethernet. Uses the same CSMA/CD

mechanism in MAC layer control. Must be compatible with 10Base-T and 100Bas-T Ethernet technology. One of the primary goals of the Gigabit Ethernet Alliance is to accelerate the Gigabit Ethernet activity. The Gigabit Ethernet standard is expected to be completed faster than other high-speed networking standard. 3. Gigabit Ethernet Characteristic Two new features have been added to CSMA/CD to enable efficient operation over a practical collision domain diameter at 1000 Mb/s. These two features, carrier extension and packet bursting, will only affect operation in Half-Duplex mode. Though minor Media Access Control (MAC) changes will be incorporated in 4. Time Issue Slot time is a critical parameter for the CSMA/CD algorithm. It is derived from the worst case round trip delay in a network, and is expressed in bit transmission times BT. In half-duplex operation mode, the relationship of slot time and the minimum frame length is very important. If the station transmit the frame smaller than the slot time, the other stations will not notice that frame when collision occurs. As a result, the minimum frame size must equal to the slot time, the worst case round trip delay. Since the signal propagation delay in a link is set by the laws of physics, any increase in data rate in a CSMA/CD network must be accompanied by either a decrease in the maximum distance spanned by the network or an increase in the slot time. The Fast Ethernet standard raised the data rate form 10Mbps to 100Mbps, the slot time was left unchanged at 512 bit times, and the maximum distance spanned by network is decreased to 205 meters. For the Gigabit Ethernet, because of the increasing of data rate, the IEEE 802.3z task force specified the slot time will increase from 512 bits to 512 bytes. 5. Carrier Extension To maintain a compatible frame format over all operating speeds is very important for Gigabit Ethernet. So, simply increasing the minimum frame size is not an acceptable solution. Carrier Extension is a simple way of supporting Gigabit Ethernet network diameters of up to 200 meters and maintain using the current CSMA/CD protocol. Carrier Extension increases the Slot Time to 512 bytes by extending shorter frames with the carrier extension symbol. The minimum frame size still remains the same as the Fast Ethernet and Standard Ethernet. Under Carrier Extension, the minimum frame size remains 512 bits the same with 10Mbps and 100Mbps networks. But the successful transmission of one frame is increasing to 512 bytes in the following way. When the length of transmitted frame is small than 512 bytes, the frame will be added a special extended carrier symbol until the length of frame equals to 512 bytes. The extended carrier symbols take place after the checksum that marks the end of the frame. But it is not a part of the frame. Carrier Extension increases the transmission time for short frames significantly, but reduces the benefit of the increased data rate. 1

In the worst case, upgrading the network connection form 100Mbps fast Ethernet to 1000Mbps Gigabit Ethernet, the minimum length 64bytes frames would allow it to send bits 10 times fast than before. Changing the minimum frame length from 512 bits to 512 bytes results in only a 25 percent net increase in throughput. 6. Frame Busting The concept of Frame Bursting is to allow stations to transmit a number of frames instead of a single frame. With Carrier Extension to the first frame in the burst the frame, the other frames are separated by a extended carrier only at least 512 bytes of the first frame can successfully transmitted. This will effectively average the wasted time (in carrier extension symbols) over the number of frames transmitted. Essentially, the first frame clears the way for the entire burst; if it has been transmitted successfully, the remainder of the frames in the burst are guaranteed not to collide in a properly designed network. The maximum burst length is based on the maximum frame length instead of the slot time. Figure1 is the structure of the Frame Burst the actual improvement is difficult and depends on the traffic patterns. Frame Bursting will be much more efficient for a traffic mix of short frames and the improvement will be less significant as the traffic mix includes longer frames. However, since short packets are part of every network, frame loss would start at a higher network load for achieving a higher network throughput. 7. Simulation and Result Analysis In real time applications Video Conference, a lot of bandwidth is needed to transmit data. The delay between each client and server, throughput and collision are the main issue. In this paper, we use the OPNET to run the simulation. The Video Conference traffic generation rate is controlled by the following conditions Conference Rate, Average Conference Duration, Frame Rate and Frame Size. In this simulation, we set the conference rate to 5 minutes, average conference duration is set to 5 minutes, frame rate is set to 20 frame per second and the frame is set to 125000 bytes. In the Gigabit Ethernet Mac layer, the burst length is set to 65536 BT FB1.0. At the simulation, we will discuss the burst length and factors effect the real time application such as Burst timer access delay, burst duration and collision. carrier sense Slot time 512 bytes 8. Result and Analysis We simulate the burst length for two Transmit data pack#1 RRRR pack#2 RR pack#3 values 65536BT FB1.0, 131072BT FB2.0 and 196608 FB3.0 in network load 15%, 30%, Figure 1 shows the structure of frame burst Frame Bursting also reduces the collision probability, since the burst of frames may collide only during the first frame. Trying to evaluate 45%, 60% and 75%. In each simulation we log the following data The Frame Burst Duration, Collision and Access Delay 2

9. The Frame Burst Duration The fame burst duration is the time counted between start to transmit the first packet to the end of the last packet within one burst. The following figures show the time of burst duration in different network load and burst length. 8 7 6 When the traffic load is low, there are few packets transmitted within one burst. Thus, the chance of collision is very similar in both burst length 65536BT and 131072BT. When the traffic load increases, there are more packets transmitted within one burst. Thus, the more packets transmitted within one burst, the more chance that other stations collide. The increasing of burst length the collision count will increase also. Time(e-5) 5 4 3 2 1 0 15% 30% 45% 60% 75% FB1.0 FB2.0 FB3.0 Figure 2. The average burst duration. According to the figures 2 above, in 15% traffic load, the packets stored in the queue of client are few, there are not many packets transmitted within one burst. So the burst duration is small. As the increasing of network load the collisions are also increasing. When the client detects the collision, packets are stored in the queue and wait for a back off delay to retransmit. When the client have attempt to transmit, it will use the burst to transmit all packets in the queue until the expiration of the burst. That is why the more traffic load, the more packets and time within one burst. In high traffic load, adjust the burst length can have more packets transmitted within one burst and more duration time. But when the network traffic goes to 75%, the burst time and the packets transmitted within one burst decreased because the collision is too high. 10. Collision Figure 3 The collision compare of traffic load 60% in burst length 0BT, 65536BT, 131072BT and 196608BT Due to the traffic load increases to 60%, there are more packets transmitted within one burst. Thus, the more packets transmitted within one burst, the more chance that other stations collide. Figure3 shows that as the increasing of burst length the collision count will increase also. As the network load increases, the collision increases also. In the high traffic load, due to the packets transmitted within one burst becomes more and more, the other station will wait more time to get change to transmit packets. As the client gets the change, it will transmit all packets in the queue until the expiration of burst. Every station has more packets stored in the queue, then it takes more time to transmit all 3

packets in the queue. As a result, the burst duration increases. Then when the bus is clear, more stations will attempt to transmit their packets. Hence, there are more collisions. When adjust the burst length, the burst duration also increases. The collision will increase with the burst length. But in the traffic load 75%, the burst duration decrease because of the collision is too high. 11. Access Delay When transmitting real time applications, the most important is the access delay. Figure 4 shows that at the burst length 131072BT has the minimum access delay when the traffic load is 65%. This is because transmitting more packets within one burst can decrease access delay. Time(ms) 350 300 250 200 150 100 50 0 FB0.5 FB1.0 FB1.5 FB2.0 FB2.5 FB3.0 45% Load 60% Load 75% Load Figure 4 The average access delay. But at the burst length 163840BT FB2.5 and 196608BT FB3.0, there are too many packets transmitted within one burst. When one station gets attempts to transmit data, it uses too long time to transmit data. For other stations, it spends more time waiting for getting chances to transmit. Then, the access delay will increase. Conclusions In our simulation, there are four traffic loads 15%, 30% 45%and 60%. As the simulation shown, in the traffic load 15%, the effect of increasing burst length has no help for network traffic. As the load increased, the effects of changing burst length are clear to be observed, especially in the high traffic load. When transmitting real time traffic, the delay of each frame is very important. It is acceptable when the frame delay is between 100-200ms. We choose the burst length 131072BT to have the minimum delay of each frame. As the network load increases, the collision increases also. In the high traffic load, due to the packets transmitted within one burst becomes more and more, the other station will have to wait more time to get chances to transmit packets. As the client gets the attempt, it will transmit all packets in the queue until the expiration of burst. Every station has more packets stored in the queue, then it takes more time to transmit all packets in the queue. As a result, the burst duration increases. After the bus is clear, a lot of stations are waiting to transmit their packets. Hence, there cause more collisions. When adjust the burst length, the burst duration also increases. The collision will increase with the burst length. In the high traffic load, when the burst length increases, the packet transmitted per burst also increases. According to the burst length, if the burst length is not expired, the station can transmit another packet behind the previous packet. The inter-frame -gap is set to 32 BT to divide each packet. The increasing of the burst length, the more packets transmitted per burst. Burst duration increases when the burst length becomes longer in high traffic load. It means each duration there are more packets to be transmitted. Therefore, the more packets to be transmitted the more duration that burst takes. To 4

calculate the duration time plus the Gigabit Ethernet transmitting rate, then we can get the data size of each duration. According to the access delay of each traffic load and each burst length, the average fame delay of traffic load 15%, 30% and 45% are smaller than 20ms, the average frame delay of traffic load 75% are more than 250ms. In the traffic load 60%, the average frame delay is between 100ms and 200ms. In transmitting real time applications it is acceptable delay. At the burst length 131072BT, the client has the minimum average frame delay 124ms when the traffic is 65%. Then in transmitting real time applications, the burst length 131072BT FB2.0 has the minimum delay. Acknowledgements This work is partly supported by the National Science Council of Ta iwan, R.O.C., under grand number NSC 87-2213-E-259-008- and NCS87-2219-E-259-001-. References 1. Mart Molle, University of California, Riverside Mohan Kalkunte and Jayant Kadambi, Advanced Micro Devices, Inc. Frame Bursting: A Technique for Scaling CSMA/CD to Gigabit Speeds IEEE Network July/August 1997 2. IEEE Draft P802.3z/D2, Media Access Control MAC Parameters, Physical Layer, Repeater and Management Parameters for 1000Mb/s Operation. 3. ANSI/IEEE Std 802.3 Carrier Sense Multiple Access with Collision Detection CSMA/CD Access Method and Physical Layer Specifications. 5 th ed, 1996 4. M. Kalkunte and J. Kadambi Packing and mtbeb Simulation Results. IEEE802.3 High-Speed Study Group plenary meeting Enschede, the Netherlands, July1996, gigabit/presentations/july1996/mksim.pdf. 5. S.Haddock, Carrier Extension Issues, IEEE 802.3 High-Speed Study Group plenary meeting, Enschede, the Netherlands, July1996, gigabit/presentations/july1996/share.txt. 6. M. Molle et al. Packet Bursting, IEEE 802.3z plenary meeting Vancouver, BC, Canada, Nov. 1996, gigabit/presentations/nov1996/mkpk_burst.pdf. 7. OPNET Modeler. MIL3 Co, Washington DC. 8. H. Frazier, Jr, Scaling CSMA/CD to 10000 Mb/s; An Update, IEEE802.3 High-Speed Study Group plenary meeting, Enschede, the Netherlands, July1996, gigabit/presentations/july1996/mksim.pdf. 9. Department of Telecommunication, Technical University of Budapest, 2 Sztoczek, Budapest, H-1111 Hungary An Ethernet compatible protocol to support real time traffic and multimedia applications. Computer Networks and ISDN Systems 29 1997 335-342 10. I. Chlamtac, An Ethernet compatible protocol for real time voice data integration Computer Networks ISDN Systems10(2) (1985) 81-86 5